Many people decide to join a gym in order to live a healthy livestyle. When they perform fitness exercises, they are often overloaded with many tasks, such as concentrating on correct exercise form, following their fitness program, counting the number of repetitions of each exercise and hydration.
In this thesis we used computer vision and machine learning for the development of a mobile web app, which takes care of one of these tasks. The app counts the number of repetitions for each exercise, while the user is performing it. The app uses a camera to record the user performing the exercise. This footage is then processed by a machine learning model that detects the body keypoints. The keypoints are used in a neural net for fitness pose estimation. A counting algorithm then uses the pose estimation to count the number of repetitions.
The app supports counting the number of repetitions for three popular fitness exercises: deadlift, squat and overhead press.
|
